All files: Update FSF address.
[gcc.git] / libgfortran / io / list_read.c
1 /* Copyright (C) 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
2 Contributed by Andy Vaught
3 Namelist input contributed by Paul Thomas
4
5 This file is part of the GNU Fortran 95 runtime library (libgfortran).
6
7 Libgfortran is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
10 any later version.
11
12 In addition to the permissions in the GNU General Public License, the
13 Free Software Foundation gives you unlimited permission to link the
14 compiled version of this file into combinations with other programs,
15 and to distribute those combinations without any restriction coming
16 from the use of this file. (The General Public License restrictions
17 do apply in other respects; for example, they cover modification of
18 the file, and distribution when not linked into a combine
19 executable.)
20
21 Libgfortran is distributed in the hope that it will be useful,
22 but WITHOUT ANY WARRANTY; without even the implied warranty of
23 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
24 GNU General Public License for more details.
25
26 You should have received a copy of the GNU General Public License
27 along with Libgfortran; see the file COPYING. If not, write to
28 the Free Software Foundation, 51 Franklin Street, Fifth Floor,
29 Boston, MA 02110-1301, USA. */
30
31
32 #include "config.h"
33 #include <string.h>
34 #include <ctype.h>
35 #include "libgfortran.h"
36 #include "io.h"
37
38
39 /* List directed input. Several parsing subroutines are practically
40 reimplemented from formatted input, the reason being that there are
41 all kinds of small differences between formatted and list directed
42 parsing. */
43
44
45 /* Subroutines for reading characters from the input. Because a
46 repeat count is ambiguous with an integer, we have to read the
47 whole digit string before seeing if there is a '*' which signals
48 the repeat count. Since we can have a lot of potential leading
49 zeros, we have to be able to back up by arbitrary amount. Because
50 the input might not be seekable, we have to buffer the data
51 ourselves. Data is buffered in scratch[] until it becomes too
52 large, after which we start allocating memory on the heap. */
53
54 static int repeat_count, saved_length, saved_used;
55 static int input_complete, at_eol, comma_flag;
56 static char last_char, *saved_string;
57 static bt saved_type;
58
59 /* A namelist specific flag used in the list directed library
60 to flag that calls are being made from namelist read (eg. to ignore
61 comments or to treat '/' as a terminator) */
62
63 static int namelist_mode;
64
65 /* A namelist specific flag used in the list directed library to flag
66 read errors and return, so that an attempt can be made to read a
67 new object name. */
68
69 static int nml_read_error;
70
71 /* Storage area for values except for strings. Must be large enough
72 to hold a complex value (two reals) of the largest kind. */
73
74 static char value[20];
75
76 #define CASE_DIGITS case '0': case '1': case '2': case '3': case '4': \
77 case '5': case '6': case '7': case '8': case '9'
78
79 #define CASE_SEPARATORS case ' ': case ',': case '/': case '\n': case '\t': \
80 case '\r'
81
82 /* This macro assumes that we're operating on a variable. */
83
84 #define is_separator(c) (c == '/' || c == ',' || c == '\n' || c == ' ' \
85 || c == '\t' || c == '\r')
86
87 /* Maximum repeat count. Less than ten times the maximum signed int32. */
88
89 #define MAX_REPEAT 200000000
90
91
92 /* Save a character to a string buffer, enlarging it as necessary. */
93
94 static void
95 push_char (char c)
96 {
97 char *new;
98
99 if (saved_string == NULL)
100 {
101 saved_string = scratch;
102 memset (saved_string,0,SCRATCH_SIZE);
103 saved_length = SCRATCH_SIZE;
104 saved_used = 0;
105 }
106
107 if (saved_used >= saved_length)
108 {
109 saved_length = 2 * saved_length;
110 new = get_mem (2 * saved_length);
111
112 memset (new,0,2 * saved_length);
113
114 memcpy (new, saved_string, saved_used);
115 if (saved_string != scratch)
116 free_mem (saved_string);
117
118 saved_string = new;
119 }
120
121 saved_string[saved_used++] = c;
122 }
123
124
125 /* Free the input buffer if necessary. */
126
127 static void
128 free_saved (void)
129 {
130 if (saved_string == NULL)
131 return;
132
133 if (saved_string != scratch)
134 free_mem (saved_string);
135
136 saved_string = NULL;
137 }
138
139
140 static char
141 next_char (void)
142 {
143 int length;
144 char c, *p;
145
146 if (last_char != '\0')
147 {
148 at_eol = 0;
149 c = last_char;
150 last_char = '\0';
151 goto done;
152 }
153
154 length = 1;
155
156 p = salloc_r (current_unit->s, &length);
157 if (p == NULL)
158 {
159 generate_error (ERROR_OS, NULL);
160 return '\0';
161 }
162
163 if (length == 0)
164 {
165 /* For internal files return a newline instead of signalling EOF. */
166 /* ??? This isn't quite right, but we don't handle internal files
167 with multiple records. */
168 if (is_internal_unit ())
169 c = '\n';
170 else
171 longjmp (g.eof_jump, 1);
172 }
173 else
174 c = *p;
175
176 done:
177 at_eol = (c == '\n' || c == '\r');
178 return c;
179 }
180
181
182 /* Push a character back onto the input. */
183
184 static void
185 unget_char (char c)
186 {
187 last_char = c;
188 }
189
190
191 /* Skip over spaces in the input. Returns the nonspace character that
192 terminated the eating and also places it back on the input. */
193
194 static char
195 eat_spaces (void)
196 {
197 char c;
198
199 do
200 {
201 c = next_char ();
202 }
203 while (c == ' ' || c == '\t');
204
205 unget_char (c);
206 return c;
207 }
208
209
210 /* Skip over a separator. Technically, we don't always eat the whole
211 separator. This is because if we've processed the last input item,
212 then a separator is unnecessary. Plus the fact that operating
213 systems usually deliver console input on a line basis.
214
215 The upshot is that if we see a newline as part of reading a
216 separator, we stop reading. If there are more input items, we
217 continue reading the separator with finish_separator() which takes
218 care of the fact that we may or may not have seen a comma as part
219 of the separator. */
220
221 static void
222 eat_separator (void)
223 {
224 char c;
225
226 eat_spaces ();
227 comma_flag = 0;
228
229 c = next_char ();
230 switch (c)
231 {
232 case ',':
233 comma_flag = 1;
234 eat_spaces ();
235 break;
236
237 case '/':
238 input_complete = 1;
239 break;
240
241 case '\n':
242 case '\r':
243 at_eol = 1;
244 break;
245
246 case '!':
247 if (namelist_mode)
248 { /* Eat a namelist comment. */
249 do
250 c = next_char ();
251 while (c != '\n');
252
253 break;
254 }
255
256 /* Fall Through... */
257
258 default:
259 unget_char (c);
260 break;
261 }
262 }
263
264
265 /* Finish processing a separator that was interrupted by a newline.
266 If we're here, then another data item is present, so we finish what
267 we started on the previous line. */
268
269 static void
270 finish_separator (void)
271 {
272 char c;
273
274 restart:
275 eat_spaces ();
276
277 c = next_char ();
278 switch (c)
279 {
280 case ',':
281 if (comma_flag)
282 unget_char (c);
283 else
284 {
285 c = eat_spaces ();
286 if (c == '\n')
287 goto restart;
288 }
289
290 break;
291
292 case '/':
293 input_complete = 1;
294 if (!namelist_mode) next_record (0);
295 break;
296
297 case '\n':
298 case '\r':
299 goto restart;
300
301 case '!':
302 if (namelist_mode)
303 {
304 do
305 c = next_char ();
306 while (c != '\n');
307
308 goto restart;
309 }
310
311 default:
312 unget_char (c);
313 break;
314 }
315 }
316
317 /* This function is needed to catch bad conversions so that namelist can
318 attempt to see if saved_string contains a new object name rather than
319 a bad value. */
320
321 static int
322 nml_bad_return (char c)
323 {
324 if (namelist_mode)
325 {
326 nml_read_error = 1;
327 unget_char(c);
328 return 1;
329 }
330 return 0;
331 }
332
333 /* Convert an unsigned string to an integer. The length value is -1
334 if we are working on a repeat count. Returns nonzero if we have a
335 range problem. As a side effect, frees the saved_string. */
336
337 static int
338 convert_integer (int length, int negative)
339 {
340 char c, *buffer, message[100];
341 int m;
342 GFC_INTEGER_LARGEST v, max, max10;
343
344 buffer = saved_string;
345 v = 0;
346
347 max = (length == -1) ? MAX_REPEAT : max_value (length, 1);
348 max10 = max / 10;
349
350 for (;;)
351 {
352 c = *buffer++;
353 if (c == '\0')
354 break;
355 c -= '0';
356
357 if (v > max10)
358 goto overflow;
359 v = 10 * v;
360
361 if (v > max - c)
362 goto overflow;
363 v += c;
364 }
365
366 m = 0;
367
368 if (length != -1)
369 {
370 if (negative)
371 v = -v;
372 set_integer (value, v, length);
373 }
374 else
375 {
376 repeat_count = v;
377
378 if (repeat_count == 0)
379 {
380 st_sprintf (message, "Zero repeat count in item %d of list input",
381 g.item_count);
382
383 generate_error (ERROR_READ_VALUE, message);
384 m = 1;
385 }
386 }
387
388 free_saved ();
389 return m;
390
391 overflow:
392 if (length == -1)
393 st_sprintf (message, "Repeat count overflow in item %d of list input",
394 g.item_count);
395 else
396 st_sprintf (message, "Integer overflow while reading item %d",
397 g.item_count);
398
399 free_saved ();
400 generate_error (ERROR_READ_VALUE, message);
401
402 return 1;
403 }
404
405
406 /* Parse a repeat count for logical and complex values which cannot
407 begin with a digit. Returns nonzero if we are done, zero if we
408 should continue on. */
409
410 static int
411 parse_repeat (void)
412 {
413 char c, message[100];
414 int repeat;
415
416 c = next_char ();
417 switch (c)
418 {
419 CASE_DIGITS:
420 repeat = c - '0';
421 break;
422
423 CASE_SEPARATORS:
424 unget_char (c);
425 eat_separator ();
426 return 1;
427
428 default:
429 unget_char (c);
430 return 0;
431 }
432
433 for (;;)
434 {
435 c = next_char ();
436 switch (c)
437 {
438 CASE_DIGITS:
439 repeat = 10 * repeat + c - '0';
440
441 if (repeat > MAX_REPEAT)
442 {
443 st_sprintf (message,
444 "Repeat count overflow in item %d of list input",
445 g.item_count);
446
447 generate_error (ERROR_READ_VALUE, message);
448 return 1;
449 }
450
451 break;
452
453 case '*':
454 if (repeat == 0)
455 {
456 st_sprintf (message,
457 "Zero repeat count in item %d of list input",
458 g.item_count);
459
460 generate_error (ERROR_READ_VALUE, message);
461 return 1;
462 }
463
464 goto done;
465
466 default:
467 goto bad_repeat;
468 }
469 }
470
471 done:
472 repeat_count = repeat;
473 return 0;
474
475 bad_repeat:
476 st_sprintf (message, "Bad repeat count in item %d of list input",
477 g.item_count);
478
479 generate_error (ERROR_READ_VALUE, message);
480 return 1;
481 }
482
483
484 /* Read a logical character on the input. */
485
486 static void
487 read_logical (int length)
488 {
489 char c, message[100];
490 int v;
491
492 if (parse_repeat ())
493 return;
494
495 c = next_char ();
496 switch (c)
497 {
498 case 't':
499 case 'T':
500 v = 1;
501 break;
502 case 'f':
503 case 'F':
504 v = 0;
505 break;
506
507 case '.':
508 c = next_char ();
509 switch (c)
510 {
511 case 't':
512 case 'T':
513 v = 1;
514 break;
515 case 'f':
516 case 'F':
517 v = 0;
518 break;
519 default:
520 goto bad_logical;
521 }
522
523 break;
524
525 CASE_SEPARATORS:
526 unget_char (c);
527 eat_separator ();
528 return; /* Null value. */
529
530 default:
531 goto bad_logical;
532 }
533
534 saved_type = BT_LOGICAL;
535 saved_length = length;
536
537 /* Eat trailing garbage. */
538 do
539 {
540 c = next_char ();
541 }
542 while (!is_separator (c));
543
544 unget_char (c);
545 eat_separator ();
546 free_saved ();
547 set_integer ((int *) value, v, length);
548
549 return;
550
551 bad_logical:
552
553 if (nml_bad_return (c))
554 return;
555
556 st_sprintf (message, "Bad logical value while reading item %d",
557 g.item_count);
558
559 generate_error (ERROR_READ_VALUE, message);
560 }
561
562
563 /* Reading integers is tricky because we can actually be reading a
564 repeat count. We have to store the characters in a buffer because
565 we could be reading an integer that is larger than the default int
566 used for repeat counts. */
567
568 static void
569 read_integer (int length)
570 {
571 char c, message[100];
572 int negative;
573
574 negative = 0;
575
576 c = next_char ();
577 switch (c)
578 {
579 case '-':
580 negative = 1;
581 /* Fall through... */
582
583 case '+':
584 c = next_char ();
585 goto get_integer;
586
587 CASE_SEPARATORS: /* Single null. */
588 unget_char (c);
589 eat_separator ();
590 return;
591
592 CASE_DIGITS:
593 push_char (c);
594 break;
595
596 default:
597 goto bad_integer;
598 }
599
600 /* Take care of what may be a repeat count. */
601
602 for (;;)
603 {
604 c = next_char ();
605 switch (c)
606 {
607 CASE_DIGITS:
608 push_char (c);
609 break;
610
611 case '*':
612 push_char ('\0');
613 goto repeat;
614
615 CASE_SEPARATORS: /* Not a repeat count. */
616 goto done;
617
618 default:
619 goto bad_integer;
620 }
621 }
622
623 repeat:
624 if (convert_integer (-1, 0))
625 return;
626
627 /* Get the real integer. */
628
629 c = next_char ();
630 switch (c)
631 {
632 CASE_DIGITS:
633 break;
634
635 CASE_SEPARATORS:
636 unget_char (c);
637 eat_separator ();
638 return;
639
640 case '-':
641 negative = 1;
642 /* Fall through... */
643
644 case '+':
645 c = next_char ();
646 break;
647 }
648
649 get_integer:
650 if (!isdigit (c))
651 goto bad_integer;
652 push_char (c);
653
654 for (;;)
655 {
656 c = next_char ();
657 switch (c)
658 {
659 CASE_DIGITS:
660 push_char (c);
661 break;
662
663 CASE_SEPARATORS:
664 goto done;
665
666 default:
667 goto bad_integer;
668 }
669 }
670
671 bad_integer:
672
673 if (nml_bad_return (c))
674 return;
675
676 free_saved ();
677
678 st_sprintf (message, "Bad integer for item %d in list input", g.item_count);
679 generate_error (ERROR_READ_VALUE, message);
680
681 return;
682
683 done:
684 unget_char (c);
685 eat_separator ();
686
687 push_char ('\0');
688 if (convert_integer (length, negative))
689 {
690 free_saved ();
691 return;
692 }
693
694 free_saved ();
695 saved_type = BT_INTEGER;
696 }
697
698
699 /* Read a character variable. */
700
701 static void
702 read_character (int length __attribute__ ((unused)))
703 {
704 char c, quote, message[100];
705
706 quote = ' '; /* Space means no quote character. */
707
708 c = next_char ();
709 switch (c)
710 {
711 CASE_DIGITS:
712 push_char (c);
713 break;
714
715 CASE_SEPARATORS:
716 unget_char (c); /* NULL value. */
717 eat_separator ();
718 return;
719
720 case '"':
721 case '\'':
722 quote = c;
723 goto get_string;
724
725 default:
726 push_char (c);
727 goto get_string;
728 }
729
730 /* Deal with a possible repeat count. */
731
732 for (;;)
733 {
734 c = next_char ();
735 switch (c)
736 {
737 CASE_DIGITS:
738 push_char (c);
739 break;
740
741 CASE_SEPARATORS:
742 unget_char (c);
743 goto done; /* String was only digits! */
744
745 case '*':
746 push_char ('\0');
747 goto got_repeat;
748
749 default:
750 push_char (c);
751 goto get_string; /* Not a repeat count after all. */
752 }
753 }
754
755 got_repeat:
756 if (convert_integer (-1, 0))
757 return;
758
759 /* Now get the real string. */
760
761 c = next_char ();
762 switch (c)
763 {
764 CASE_SEPARATORS:
765 unget_char (c); /* Repeated NULL values. */
766 eat_separator ();
767 return;
768
769 case '"':
770 case '\'':
771 quote = c;
772 break;
773
774 default:
775 push_char (c);
776 break;
777 }
778
779 get_string:
780 for (;;)
781 {
782 c = next_char ();
783 switch (c)
784 {
785 case '"':
786 case '\'':
787 if (c != quote)
788 {
789 push_char (c);
790 break;
791 }
792
793 /* See if we have a doubled quote character or the end of
794 the string. */
795
796 c = next_char ();
797 if (c == quote)
798 {
799 push_char (quote);
800 break;
801 }
802
803 unget_char (c);
804 goto done;
805
806 CASE_SEPARATORS:
807 if (quote == ' ')
808 {
809 unget_char (c);
810 goto done;
811 }
812
813 if (c != '\n')
814 push_char (c);
815 break;
816
817 default:
818 push_char (c);
819 break;
820 }
821 }
822
823 /* At this point, we have to have a separator, or else the string is
824 invalid. */
825 done:
826 c = next_char ();
827 if (is_separator (c))
828 {
829 unget_char (c);
830 eat_separator ();
831 saved_type = BT_CHARACTER;
832 }
833 else
834 {
835 free_saved ();
836 st_sprintf (message, "Invalid string input in item %d", g.item_count);
837 generate_error (ERROR_READ_VALUE, message);
838 }
839 }
840
841
842 /* Parse a component of a complex constant or a real number that we
843 are sure is already there. This is a straight real number parser. */
844
845 static int
846 parse_real (void *buffer, int length)
847 {
848 char c, message[100];
849 int m, seen_dp;
850
851 c = next_char ();
852 if (c == '-' || c == '+')
853 {
854 push_char (c);
855 c = next_char ();
856 }
857
858 if (!isdigit (c) && c != '.')
859 goto bad;
860
861 push_char (c);
862
863 seen_dp = (c == '.') ? 1 : 0;
864
865 for (;;)
866 {
867 c = next_char ();
868 switch (c)
869 {
870 CASE_DIGITS:
871 push_char (c);
872 break;
873
874 case '.':
875 if (seen_dp)
876 goto bad;
877
878 seen_dp = 1;
879 push_char (c);
880 break;
881
882 case 'e':
883 case 'E':
884 case 'd':
885 case 'D':
886 push_char ('e');
887 goto exp1;
888
889 case '-':
890 case '+':
891 push_char ('e');
892 push_char (c);
893 c = next_char ();
894 goto exp2;
895
896 CASE_SEPARATORS:
897 unget_char (c);
898 goto done;
899
900 default:
901 goto done;
902 }
903 }
904
905 exp1:
906 c = next_char ();
907 if (c != '-' && c != '+')
908 push_char ('+');
909 else
910 {
911 push_char (c);
912 c = next_char ();
913 }
914
915 exp2:
916 if (!isdigit (c))
917 goto bad;
918 push_char (c);
919
920 for (;;)
921 {
922 c = next_char ();
923 switch (c)
924 {
925 CASE_DIGITS:
926 push_char (c);
927 break;
928
929 CASE_SEPARATORS:
930 unget_char (c);
931 goto done;
932
933 default:
934 goto done;
935 }
936 }
937
938 done:
939 unget_char (c);
940 push_char ('\0');
941
942 m = convert_real (buffer, saved_string, length);
943 free_saved ();
944
945 return m;
946
947 bad:
948 free_saved ();
949 st_sprintf (message, "Bad floating point number for item %d", g.item_count);
950 generate_error (ERROR_READ_VALUE, message);
951
952 return 1;
953 }
954
955
956 /* Reading a complex number is straightforward because we can tell
957 what it is right away. */
958
959 static void
960 read_complex (int length)
961 {
962 char message[100];
963 char c;
964
965 if (parse_repeat ())
966 return;
967
968 c = next_char ();
969 switch (c)
970 {
971 case '(':
972 break;
973
974 CASE_SEPARATORS:
975 unget_char (c);
976 eat_separator ();
977 return;
978
979 default:
980 goto bad_complex;
981 }
982
983 eat_spaces ();
984 if (parse_real (value, length))
985 return;
986
987 eol_1:
988 eat_spaces ();
989 c = next_char ();
990 if (c == '\n' || c== '\r')
991 goto eol_1;
992 else
993 unget_char (c);
994
995 if (next_char () != ',')
996 goto bad_complex;
997
998 eol_2:
999 eat_spaces ();
1000 c = next_char ();
1001 if (c == '\n' || c== '\r')
1002 goto eol_2;
1003 else
1004 unget_char (c);
1005
1006 if (parse_real (value + length, length))
1007 return;
1008
1009 eat_spaces ();
1010 if (next_char () != ')')
1011 goto bad_complex;
1012
1013 c = next_char ();
1014 if (!is_separator (c))
1015 goto bad_complex;
1016
1017 unget_char (c);
1018 eat_separator ();
1019
1020 free_saved ();
1021 saved_type = BT_COMPLEX;
1022 return;
1023
1024 bad_complex:
1025
1026 if (nml_bad_return (c))
1027 return;
1028
1029 st_sprintf (message, "Bad complex value in item %d of list input",
1030 g.item_count);
1031
1032 generate_error (ERROR_READ_VALUE, message);
1033 }
1034
1035
1036 /* Parse a real number with a possible repeat count. */
1037
1038 static void
1039 read_real (int length)
1040 {
1041 char c, message[100];
1042 int seen_dp;
1043
1044 seen_dp = 0;
1045
1046 c = next_char ();
1047 switch (c)
1048 {
1049 CASE_DIGITS:
1050 push_char (c);
1051 break;
1052
1053 case '.':
1054 push_char (c);
1055 seen_dp = 1;
1056 break;
1057
1058 case '+':
1059 case '-':
1060 goto got_sign;
1061
1062 CASE_SEPARATORS:
1063 unget_char (c); /* Single null. */
1064 eat_separator ();
1065 return;
1066
1067 default:
1068 goto bad_real;
1069 }
1070
1071 /* Get the digit string that might be a repeat count. */
1072
1073 for (;;)
1074 {
1075 c = next_char ();
1076 switch (c)
1077 {
1078 CASE_DIGITS:
1079 push_char (c);
1080 break;
1081
1082 case '.':
1083 if (seen_dp)
1084 goto bad_real;
1085
1086 seen_dp = 1;
1087 push_char (c);
1088 goto real_loop;
1089
1090 case 'E':
1091 case 'e':
1092 case 'D':
1093 case 'd':
1094 goto exp1;
1095
1096 case '+':
1097 case '-':
1098 push_char ('e');
1099 push_char (c);
1100 c = next_char ();
1101 goto exp2;
1102
1103 case '*':
1104 push_char ('\0');
1105 goto got_repeat;
1106
1107 CASE_SEPARATORS:
1108 if (c != '\n' && c != ',' && c != '\r')
1109 unget_char (c);
1110 goto done;
1111
1112 default:
1113 goto bad_real;
1114 }
1115 }
1116
1117 got_repeat:
1118 if (convert_integer (-1, 0))
1119 return;
1120
1121 /* Now get the number itself. */
1122
1123 c = next_char ();
1124 if (is_separator (c))
1125 { /* Repeated null value. */
1126 unget_char (c);
1127 eat_separator ();
1128 return;
1129 }
1130
1131 if (c != '-' && c != '+')
1132 push_char ('+');
1133 else
1134 {
1135 got_sign:
1136 push_char (c);
1137 c = next_char ();
1138 }
1139
1140 if (!isdigit (c) && c != '.')
1141 goto bad_real;
1142
1143 if (c == '.')
1144 {
1145 if (seen_dp)
1146 goto bad_real;
1147 else
1148 seen_dp = 1;
1149 }
1150
1151 push_char (c);
1152
1153 real_loop:
1154 for (;;)
1155 {
1156 c = next_char ();
1157 switch (c)
1158 {
1159 CASE_DIGITS:
1160 push_char (c);
1161 break;
1162
1163 CASE_SEPARATORS:
1164 goto done;
1165
1166 case '.':
1167 if (seen_dp)
1168 goto bad_real;
1169
1170 seen_dp = 1;
1171 push_char (c);
1172 break;
1173
1174 case 'E':
1175 case 'e':
1176 case 'D':
1177 case 'd':
1178 goto exp1;
1179
1180 case '+':
1181 case '-':
1182 push_char ('e');
1183 push_char (c);
1184 c = next_char ();
1185 goto exp2;
1186
1187 default:
1188 goto bad_real;
1189 }
1190 }
1191
1192 exp1:
1193 push_char ('e');
1194
1195 c = next_char ();
1196 if (c != '+' && c != '-')
1197 push_char ('+');
1198 else
1199 {
1200 push_char (c);
1201 c = next_char ();
1202 }
1203
1204 exp2:
1205 if (!isdigit (c))
1206 goto bad_real;
1207 push_char (c);
1208
1209 for (;;)
1210 {
1211 c = next_char ();
1212
1213 switch (c)
1214 {
1215 CASE_DIGITS:
1216 push_char (c);
1217 break;
1218
1219 CASE_SEPARATORS:
1220 goto done;
1221
1222 default:
1223 goto bad_real;
1224 }
1225 }
1226
1227 done:
1228 unget_char (c);
1229 eat_separator ();
1230 push_char ('\0');
1231 if (convert_real (value, saved_string, length))
1232 return;
1233
1234 free_saved ();
1235 saved_type = BT_REAL;
1236 return;
1237
1238 bad_real:
1239
1240 if (nml_bad_return (c))
1241 return;
1242
1243 st_sprintf (message, "Bad real number in item %d of list input",
1244 g.item_count);
1245
1246 generate_error (ERROR_READ_VALUE, message);
1247 }
1248
1249
1250 /* Check the current type against the saved type to make sure they are
1251 compatible. Returns nonzero if incompatible. */
1252
1253 static int
1254 check_type (bt type, int len)
1255 {
1256 char message[100];
1257
1258 if (saved_type != BT_NULL && saved_type != type)
1259 {
1260 st_sprintf (message, "Read type %s where %s was expected for item %d",
1261 type_name (saved_type), type_name (type), g.item_count);
1262
1263 generate_error (ERROR_READ_VALUE, message);
1264 return 1;
1265 }
1266
1267 if (saved_type == BT_NULL || saved_type == BT_CHARACTER)
1268 return 0;
1269
1270 if (saved_length != len)
1271 {
1272 st_sprintf (message,
1273 "Read kind %d %s where kind %d is required for item %d",
1274 saved_length, type_name (saved_type), len, g.item_count);
1275 generate_error (ERROR_READ_VALUE, message);
1276 return 1;
1277 }
1278
1279 return 0;
1280 }
1281
1282
1283 /* Top level data transfer subroutine for list reads. Because we have
1284 to deal with repeat counts, the data item is always saved after
1285 reading, usually in the value[] array. If a repeat count is
1286 greater than one, we copy the data item multiple times. */
1287
1288 void
1289 list_formatted_read (bt type, void *p, int len)
1290 {
1291 char c;
1292 int m;
1293
1294 namelist_mode = 0;
1295
1296 if (setjmp (g.eof_jump))
1297 {
1298 generate_error (ERROR_END, NULL);
1299 return;
1300 }
1301
1302 if (g.first_item)
1303 {
1304 g.first_item = 0;
1305 input_complete = 0;
1306 repeat_count = 1;
1307 at_eol = 0;
1308
1309 c = eat_spaces ();
1310 if (is_separator (c))
1311 { /* Found a null value. */
1312 eat_separator ();
1313 repeat_count = 0;
1314 if (at_eol)
1315 finish_separator ();
1316 else
1317 return;
1318 }
1319
1320 }
1321 else
1322 {
1323 if (input_complete)
1324 return;
1325
1326 if (repeat_count > 0)
1327 {
1328 if (check_type (type, len))
1329 return;
1330 goto set_value;
1331 }
1332
1333 if (at_eol)
1334 finish_separator ();
1335 else
1336 {
1337 eat_spaces ();
1338 /* trailing spaces prior to end of line */
1339 if (at_eol)
1340 finish_separator ();
1341 }
1342
1343 saved_type = BT_NULL;
1344 repeat_count = 1;
1345 }
1346
1347 switch (type)
1348 {
1349 case BT_INTEGER:
1350 read_integer (len);
1351 break;
1352 case BT_LOGICAL:
1353 read_logical (len);
1354 break;
1355 case BT_CHARACTER:
1356 read_character (len);
1357 break;
1358 case BT_REAL:
1359 read_real (len);
1360 break;
1361 case BT_COMPLEX:
1362 read_complex (len);
1363 break;
1364 default:
1365 internal_error ("Bad type for list read");
1366 }
1367
1368 if (saved_type != BT_CHARACTER && saved_type != BT_NULL)
1369 saved_length = len;
1370
1371 if (ioparm.library_return != LIBRARY_OK)
1372 return;
1373
1374 set_value:
1375 switch (saved_type)
1376 {
1377 case BT_COMPLEX:
1378 len = 2 * len;
1379 /* Fall through. */
1380
1381 case BT_INTEGER:
1382 case BT_REAL:
1383 case BT_LOGICAL:
1384 memcpy (p, value, len);
1385 break;
1386
1387 case BT_CHARACTER:
1388 if (saved_string)
1389 {
1390 m = (len < saved_used) ? len : saved_used;
1391 memcpy (p, saved_string, m);
1392 }
1393 else
1394 /* Just delimiters encountered, nothing to copy but SPACE. */
1395 m = 0;
1396
1397 if (m < len)
1398 memset (((char *) p) + m, ' ', len - m);
1399 break;
1400
1401 case BT_NULL:
1402 break;
1403 }
1404
1405 if (--repeat_count <= 0)
1406 free_saved ();
1407 }
1408
1409 void
1410 init_at_eol(void)
1411 {
1412 at_eol = 0;
1413 }
1414
1415 /* Finish a list read. */
1416
1417 void
1418 finish_list_read (void)
1419 {
1420 char c;
1421
1422 free_saved ();
1423
1424 if (at_eol)
1425 {
1426 at_eol = 0;
1427 return;
1428 }
1429
1430 do
1431 {
1432 c = next_char ();
1433 }
1434 while (c != '\n');
1435 }
1436
1437 /* NAMELIST INPUT
1438
1439 void namelist_read (void)
1440 calls:
1441 static void nml_match_name (char *name, int len)
1442 static int nml_query (void)
1443 static int nml_get_obj_data (void)
1444 calls:
1445 static void nml_untouch_nodes (void)
1446 static namelist_info * find_nml_node (char * var_name)
1447 static int nml_parse_qualifier(descriptor_dimension * ad,
1448 nml_loop_spec * ls, int rank)
1449 static void nml_touch_nodes (namelist_info * nl)
1450 static int nml_read_obj (namelist_info * nl, index_type offset)
1451 calls:
1452 -itself- */
1453
1454 /* Carries error messages from the qualifier parser. */
1455 static char parse_err_msg[30];
1456
1457 /* Carries error messages for error returns. */
1458 static char nml_err_msg[100];
1459
1460 /* Pointer to the previously read object, in case attempt is made to read
1461 new object name. Should this fail, error message can give previous
1462 name. */
1463
1464 static namelist_info * prev_nl;
1465
1466 /* Lower index for substring qualifier. */
1467
1468 static index_type clow;
1469
1470 /* Upper index for substring qualifier. */
1471
1472 static index_type chigh;
1473
1474 /* Inputs a rank-dimensional qualifier, which can contain
1475 singlets, doublets, triplets or ':' with the standard meanings. */
1476
1477 static try
1478 nml_parse_qualifier(descriptor_dimension * ad,
1479 nml_loop_spec * ls, int rank)
1480 {
1481 int dim;
1482 int indx;
1483 int neg;
1484 int null_flag;
1485 char c;
1486
1487 /* The next character in the stream should be the '('. */
1488
1489 c = next_char ();
1490
1491 /* Process the qualifier, by dimension and triplet. */
1492
1493 for (dim=0; dim < rank; dim++ )
1494 {
1495 for (indx=0; indx<3; indx++)
1496 {
1497 free_saved ();
1498 eat_spaces ();
1499 neg = 0;
1500
1501 /*process a potential sign. */
1502
1503 c = next_char ();
1504 switch (c)
1505 {
1506 case '-':
1507 neg = 1;
1508 break;
1509
1510 case '+':
1511 break;
1512
1513 default:
1514 unget_char (c);
1515 break;
1516 }
1517
1518 /*process characters up to the next ':' , ',' or ')' */
1519
1520 for (;;)
1521 {
1522 c = next_char ();
1523
1524 switch (c)
1525 {
1526 case ':':
1527 break;
1528
1529 case ',': case ')':
1530 if ( (c==',' && dim == rank -1)
1531 || (c==')' && dim < rank -1))
1532 {
1533 st_sprintf (parse_err_msg,
1534 "Bad number of index fields");
1535 goto err_ret;
1536 }
1537 break;
1538
1539 CASE_DIGITS:
1540 push_char (c);
1541 continue;
1542
1543 case ' ': case '\t':
1544 eat_spaces ();
1545 c = next_char ();
1546 break;
1547
1548 default:
1549 st_sprintf (parse_err_msg, "Bad character in index");
1550 goto err_ret;
1551 }
1552
1553 if (( c==',' || c==')') && indx==0 && saved_string == 0 )
1554 {
1555 st_sprintf (parse_err_msg, "Null index field");
1556 goto err_ret;
1557 }
1558
1559 if ( ( c==':' && indx==1 && saved_string == 0)
1560 || (indx==2 && saved_string == 0))
1561 {
1562 st_sprintf(parse_err_msg, "Bad index triplet");
1563 goto err_ret;
1564 }
1565
1566 /* If '( : ? )' or '( ? : )' break and flag read failure. */
1567 null_flag = 0;
1568 if ( (c==':' && indx==0 && saved_string == 0)
1569 || (indx==1 && saved_string == 0))
1570 {
1571 null_flag = 1;
1572 break;
1573 }
1574
1575 /* Now read the index. */
1576
1577 if (convert_integer (sizeof(int),neg))
1578 {
1579 st_sprintf (parse_err_msg, "Bad integer in index");
1580 goto err_ret;
1581 }
1582 break;
1583 }
1584
1585 /*feed the index values to the triplet arrays. */
1586
1587 if (!null_flag)
1588 {
1589 if (indx == 0)
1590 ls[dim].start = *(int *)value;
1591 if (indx == 1)
1592 ls[dim].end = *(int *)value;
1593 if (indx == 2)
1594 ls[dim].step = *(int *)value;
1595 }
1596
1597 /*singlet or doublet indices */
1598
1599 if (c==',' || c==')')
1600 {
1601 if (indx == 0)
1602 {
1603 ls[dim].start = *(int *)value;
1604 ls[dim].end = *(int *)value;
1605 }
1606 break;
1607 }
1608 }
1609
1610 /*Check the values of the triplet indices. */
1611
1612 if ( (ls[dim].start > (ssize_t)ad[dim].ubound)
1613 || (ls[dim].start < (ssize_t)ad[dim].lbound)
1614 || (ls[dim].end > (ssize_t)ad[dim].ubound)
1615 || (ls[dim].end < (ssize_t)ad[dim].lbound))
1616 {
1617 st_sprintf (parse_err_msg, "Index %d out of range", dim + 1);
1618 goto err_ret;
1619 }
1620 if (((ls[dim].end - ls[dim].start ) * ls[dim].step < 0)
1621 || (ls[dim].step == 0))
1622 {
1623 st_sprintf (parse_err_msg, "Bad range in index %d", dim + 1);
1624 goto err_ret;
1625 }
1626
1627 /* Initialise the loop index counter. */
1628
1629 ls[dim].idx = ls[dim].start;
1630
1631 }
1632 eat_spaces ();
1633 return SUCCESS;
1634
1635 err_ret:
1636
1637 return FAILURE;
1638 }
1639
1640 static namelist_info *
1641 find_nml_node (char * var_name)
1642 {
1643 namelist_info * t = ionml;
1644 while (t != NULL)
1645 {
1646 if (strcmp (var_name,t->var_name) == 0)
1647 {
1648 t->touched = 1;
1649 return t;
1650 }
1651 t = t->next;
1652 }
1653 return NULL;
1654 }
1655
1656 /* Visits all the components of a derived type that have
1657 not explicitly been identified in the namelist input.
1658 touched is set and the loop specification initialised
1659 to default values */
1660
1661 static void
1662 nml_touch_nodes (namelist_info * nl)
1663 {
1664 index_type len = strlen (nl->var_name) + 1;
1665 int dim;
1666 char * ext_name = (char*)get_mem (len + 1);
1667 strcpy (ext_name, nl->var_name);
1668 strcat (ext_name, "%");
1669 for (nl = nl->next; nl; nl = nl->next)
1670 {
1671 if (strncmp (nl->var_name, ext_name, len) == 0)
1672 {
1673 nl->touched = 1;
1674 for (dim=0; dim < nl->var_rank; dim++)
1675 {
1676 nl->ls[dim].step = 1;
1677 nl->ls[dim].end = nl->dim[dim].ubound;
1678 nl->ls[dim].start = nl->dim[dim].lbound;
1679 nl->ls[dim].idx = nl->ls[dim].start;
1680 }
1681 }
1682 else
1683 break;
1684 }
1685 free_mem (ext_name);
1686 return;
1687 }
1688
1689 /* Resets touched for the entire list of nml_nodes, ready for a
1690 new object. */
1691
1692 static void
1693 nml_untouch_nodes (void)
1694 {
1695 namelist_info * t;
1696 for (t = ionml; t; t = t->next)
1697 t->touched = 0;
1698 return;
1699 }
1700
1701 /* Attempts to input name to namelist name. Returns nml_read_error = 1
1702 on no match. */
1703
1704 static void
1705 nml_match_name (const char *name, index_type len)
1706 {
1707 index_type i;
1708 char c;
1709 nml_read_error = 0;
1710 for (i = 0; i < len; i++)
1711 {
1712 c = next_char ();
1713 if (tolower (c) != tolower (name[i]))
1714 {
1715 nml_read_error = 1;
1716 break;
1717 }
1718 }
1719 }
1720
1721 /* If the namelist read is from stdin, output the current state of the
1722 namelist to stdout. This is used to implement the non-standard query
1723 features, ? and =?. If c == '=' the full namelist is printed. Otherwise
1724 the names alone are printed. */
1725
1726 static void
1727 nml_query (char c)
1728 {
1729 gfc_unit * temp_unit;
1730 namelist_info * nl;
1731 index_type len;
1732 char * p;
1733
1734 if (current_unit->unit_number != options.stdin_unit)
1735 return;
1736
1737 /* Store the current unit and transfer to stdout. */
1738
1739 temp_unit = current_unit;
1740 current_unit = find_unit (options.stdout_unit);
1741
1742 if (current_unit)
1743 {
1744 g.mode =WRITING;
1745 next_record (0);
1746
1747 /* Write the namelist in its entirety. */
1748
1749 if (c == '=')
1750 namelist_write ();
1751
1752 /* Or write the list of names. */
1753
1754 else
1755 {
1756
1757 /* "&namelist_name\n" */
1758
1759 len = ioparm.namelist_name_len;
1760 p = write_block (len + 2);
1761 if (!p)
1762 goto query_return;
1763 memcpy (p, "&", 1);
1764 memcpy ((char*)(p + 1), ioparm.namelist_name, len);
1765 memcpy ((char*)(p + len + 1), "\n", 1);
1766 for (nl =ionml; nl; nl = nl->next)
1767 {
1768
1769 /* " var_name\n" */
1770
1771 len = strlen (nl->var_name);
1772 p = write_block (len + 2);
1773 if (!p)
1774 goto query_return;
1775 memcpy (p, " ", 1);
1776 memcpy ((char*)(p + 1), nl->var_name, len);
1777 memcpy ((char*)(p + len + 1), "\n", 1);
1778 }
1779
1780 /* "&end\n" */
1781
1782 p = write_block (5);
1783 if (!p)
1784 goto query_return;
1785 memcpy (p, "&end\n", 5);
1786 }
1787
1788 /* Flush the stream to force immediate output. */
1789
1790 flush (current_unit->s);
1791 }
1792
1793 query_return:
1794
1795 /* Restore the current unit. */
1796
1797 current_unit = temp_unit;
1798 g.mode = READING;
1799 return;
1800 }
1801
1802 /* Reads and stores the input for the namelist object nl. For an array,
1803 the function loops over the ranges defined by the loop specification.
1804 This default to all the data or to the specification from a qualifier.
1805 nml_read_obj recursively calls itself to read derived types. It visits
1806 all its own components but only reads data for those that were touched
1807 when the name was parsed. If a read error is encountered, an attempt is
1808 made to return to read a new object name because the standard allows too
1809 little data to be available. On the other hand, too much data is an
1810 error. */
1811
1812 static try
1813 nml_read_obj (namelist_info * nl, index_type offset)
1814 {
1815
1816 namelist_info * cmp;
1817 char * obj_name;
1818 int nml_carry;
1819 int len;
1820 int dim;
1821 index_type dlen;
1822 index_type m;
1823 index_type obj_name_len;
1824 void * pdata ;
1825
1826 /* This object not touched in name parsing. */
1827
1828 if (!nl->touched)
1829 return SUCCESS;
1830
1831 repeat_count = 0;
1832 eat_spaces();
1833
1834 len = nl->len;
1835 switch (nl->type)
1836 {
1837
1838 case GFC_DTYPE_INTEGER:
1839 case GFC_DTYPE_LOGICAL:
1840 case GFC_DTYPE_REAL:
1841 dlen = len;
1842 break;
1843
1844 case GFC_DTYPE_COMPLEX:
1845 dlen = 2* len;
1846 break;
1847
1848 case GFC_DTYPE_CHARACTER:
1849 dlen = chigh ? (chigh - clow + 1) : nl->string_length;
1850 break;
1851
1852 default:
1853 dlen = 0;
1854 }
1855
1856 do
1857 {
1858
1859 /* Update the pointer to the data, using the current index vector */
1860
1861 pdata = (void*)(nl->mem_pos + offset);
1862 for (dim = 0; dim < nl->var_rank; dim++)
1863 pdata = (void*)(pdata + (nl->ls[dim].idx - nl->dim[dim].lbound) *
1864 nl->dim[dim].stride * nl->size);
1865
1866 /* Reset the error flag and try to read next value, if
1867 repeat_count=0 */
1868
1869 nml_read_error = 0;
1870 nml_carry = 0;
1871 if (--repeat_count <= 0)
1872 {
1873 if (input_complete)
1874 return SUCCESS;
1875 if (at_eol)
1876 finish_separator ();
1877 if (input_complete)
1878 return SUCCESS;
1879
1880 /* GFC_TYPE_UNKNOWN through for nulls and is detected
1881 after the switch block. */
1882
1883 saved_type = GFC_DTYPE_UNKNOWN;
1884 free_saved ();
1885
1886 switch (nl->type)
1887 {
1888 case GFC_DTYPE_INTEGER:
1889 read_integer (len);
1890 break;
1891
1892 case GFC_DTYPE_LOGICAL:
1893 read_logical (len);
1894 break;
1895
1896 case GFC_DTYPE_CHARACTER:
1897 read_character (len);
1898 break;
1899
1900 case GFC_DTYPE_REAL:
1901 read_real (len);
1902 break;
1903
1904 case GFC_DTYPE_COMPLEX:
1905 read_complex (len);
1906 break;
1907
1908 case GFC_DTYPE_DERIVED:
1909 obj_name_len = strlen (nl->var_name) + 1;
1910 obj_name = get_mem (obj_name_len+1);
1911 strcpy (obj_name, nl->var_name);
1912 strcat (obj_name, "%");
1913
1914 /* Now loop over the components. Update the component pointer
1915 with the return value from nml_write_obj. This loop jumps
1916 past nested derived types by testing if the potential
1917 component name contains '%'. */
1918
1919 for (cmp = nl->next;
1920 cmp &&
1921 !strncmp (cmp->var_name, obj_name, obj_name_len) &&
1922 !strchr (cmp->var_name + obj_name_len, '%');
1923 cmp = cmp->next)
1924 {
1925
1926 if (nml_read_obj (cmp, (index_type)(pdata - nl->mem_pos)) == FAILURE)
1927 {
1928 free_mem (obj_name);
1929 return FAILURE;
1930 }
1931
1932 if (input_complete)
1933 {
1934 free_mem (obj_name);
1935 return SUCCESS;
1936 }
1937 }
1938
1939 free_mem (obj_name);
1940 goto incr_idx;
1941
1942 default:
1943 st_sprintf (nml_err_msg, "Bad type for namelist object %s",
1944 nl->var_name );
1945 internal_error (nml_err_msg);
1946 goto nml_err_ret;
1947 }
1948 }
1949
1950 /* The standard permits array data to stop short of the number of
1951 elements specified in the loop specification. In this case, we
1952 should be here with nml_read_error != 0. Control returns to
1953 nml_get_obj_data and an attempt is made to read object name. */
1954
1955 prev_nl = nl;
1956 if (nml_read_error)
1957 return SUCCESS;
1958
1959 if (saved_type == GFC_DTYPE_UNKNOWN)
1960 goto incr_idx;
1961
1962
1963 /* Note the switch from GFC_DTYPE_type to BT_type at this point.
1964 This comes about because the read functions return BT_types. */
1965
1966 switch (saved_type)
1967 {
1968
1969 case BT_COMPLEX:
1970 case BT_REAL:
1971 case BT_INTEGER:
1972 case BT_LOGICAL:
1973 memcpy (pdata, value, dlen);
1974 break;
1975
1976 case BT_CHARACTER:
1977 m = (dlen < saved_used) ? dlen : saved_used;
1978 pdata = (void*)( pdata + clow - 1 );
1979 memcpy (pdata, saved_string, m);
1980 if (m < dlen)
1981 memset ((void*)( pdata + m ), ' ', dlen - m);
1982 break;
1983
1984 default:
1985 break;
1986 }
1987
1988 /* Break out of loop if scalar. */
1989
1990 if (!nl->var_rank)
1991 break;
1992
1993 /* Now increment the index vector. */
1994
1995 incr_idx:
1996
1997 nml_carry = 1;
1998 for (dim = 0; dim < nl->var_rank; dim++)
1999 {
2000 nl->ls[dim].idx += nml_carry * nl->ls[dim].step;
2001 nml_carry = 0;
2002 if (((nl->ls[dim].step > 0) && (nl->ls[dim].idx > nl->ls[dim].end))
2003 ||
2004 ((nl->ls[dim].step < 0) && (nl->ls[dim].idx < nl->ls[dim].end)))
2005 {
2006 nl->ls[dim].idx = nl->ls[dim].start;
2007 nml_carry = 1;
2008 }
2009 }
2010 } while (!nml_carry);
2011
2012 if (repeat_count > 1)
2013 {
2014 st_sprintf (nml_err_msg, "Repeat count too large for namelist object %s" ,
2015 nl->var_name );
2016 goto nml_err_ret;
2017 }
2018 return SUCCESS;
2019
2020 nml_err_ret:
2021
2022 return FAILURE;
2023 }
2024
2025 /* Parses the object name, including array and substring qualifiers. It
2026 iterates over derived type components, touching those components and
2027 setting their loop specifications, if there is a qualifier. If the
2028 object is itself a derived type, its components and subcomponents are
2029 touched. nml_read_obj is called at the end and this reads the data in
2030 the manner specified by the object name. */
2031
2032 static try
2033 nml_get_obj_data (void)
2034 {
2035 char c;
2036 char * ext_name;
2037 namelist_info * nl;
2038 namelist_info * first_nl = NULL;
2039 namelist_info * root_nl = NULL;
2040 int dim;
2041 int component_flag;
2042
2043 /* Look for end of input or object name. If '?' or '=?' are encountered
2044 in stdin, print the node names or the namelist to stdout. */
2045
2046 eat_separator ();
2047 if (input_complete)
2048 return SUCCESS;
2049
2050 if ( at_eol )
2051 finish_separator ();
2052 if (input_complete)
2053 return SUCCESS;
2054
2055 c = next_char ();
2056 switch (c)
2057 {
2058 case '=':
2059 c = next_char ();
2060 if (c != '?')
2061 {
2062 st_sprintf (nml_err_msg, "namelist read: missplaced = sign");
2063 goto nml_err_ret;
2064 }
2065 nml_query ('=');
2066 return SUCCESS;
2067
2068 case '?':
2069 nml_query ('?');
2070 return SUCCESS;
2071
2072 case '$':
2073 case '&':
2074 nml_match_name ("end", 3);
2075 if (nml_read_error)
2076 {
2077 st_sprintf (nml_err_msg, "namelist not terminated with / or &end");
2078 goto nml_err_ret;
2079 }
2080 case '/':
2081 input_complete = 1;
2082 return SUCCESS;
2083
2084 default :
2085 break;
2086 }
2087
2088 /* Untouch all nodes of the namelist and reset the flag that is set for
2089 derived type components. */
2090
2091 nml_untouch_nodes();
2092 component_flag = 0;
2093
2094 /* Get the object name - should '!' and '\n' be permitted separators? */
2095
2096 get_name:
2097
2098 free_saved ();
2099
2100 do
2101 {
2102 push_char(tolower(c));
2103 c = next_char ();
2104 } while (!( c=='=' || c==' ' || c=='\t' || c =='(' || c =='%' ));
2105
2106 unget_char (c);
2107
2108 /* Check that the name is in the namelist and get pointer to object.
2109 Three error conditions exist: (i) An attempt is being made to
2110 identify a non-existent object, following a failed data read or
2111 (ii) The object name does not exist or (iii) Too many data items
2112 are present for an object. (iii) gives the same error message
2113 as (i) */
2114
2115 push_char ('\0');
2116
2117 if (component_flag)
2118 {
2119 ext_name = (char*)get_mem (strlen (root_nl->var_name)
2120 + (saved_string ? strlen (saved_string) : 0)
2121 + 1);
2122 strcpy (ext_name, root_nl->var_name);
2123 strcat (ext_name, saved_string);
2124 nl = find_nml_node (ext_name);
2125 free_mem (ext_name);
2126 }
2127 else
2128 nl = find_nml_node (saved_string);
2129
2130 if (nl == NULL)
2131 {
2132 if (nml_read_error && prev_nl)
2133 st_sprintf (nml_err_msg, "Bad data for namelist object %s",
2134 prev_nl->var_name);
2135
2136 else
2137 st_sprintf (nml_err_msg, "Cannot match namelist object name %s",
2138 saved_string);
2139
2140 goto nml_err_ret;
2141 }
2142
2143 /* Get the length, data length, base pointer and rank of the variable.
2144 Set the default loop specification first. */
2145
2146 for (dim=0; dim < nl->var_rank; dim++)
2147 {
2148 nl->ls[dim].step = 1;
2149 nl->ls[dim].end = nl->dim[dim].ubound;
2150 nl->ls[dim].start = nl->dim[dim].lbound;
2151 nl->ls[dim].idx = nl->ls[dim].start;
2152 }
2153
2154 /* Check to see if there is a qualifier: if so, parse it.*/
2155
2156 if (c == '(' && nl->var_rank)
2157 {
2158 if (nml_parse_qualifier (nl->dim, nl->ls, nl->var_rank) == FAILURE)
2159 {
2160 st_sprintf (nml_err_msg, "%s for namelist variable %s",
2161 parse_err_msg, nl->var_name);
2162 goto nml_err_ret;
2163 }
2164 c = next_char ();
2165 unget_char (c);
2166 }
2167
2168 /* Now parse a derived type component. The root namelist_info address
2169 is backed up, as is the previous component level. The component flag
2170 is set and the iteration is made by jumping back to get_name. */
2171
2172 if (c == '%')
2173 {
2174
2175 if (nl->type != GFC_DTYPE_DERIVED)
2176 {
2177 st_sprintf (nml_err_msg, "Attempt to get derived component for %s",
2178 nl->var_name);
2179 goto nml_err_ret;
2180 }
2181
2182 if (!component_flag)
2183 first_nl = nl;
2184
2185 root_nl = nl;
2186 component_flag = 1;
2187 c = next_char ();
2188 goto get_name;
2189
2190 }
2191
2192 /* Parse a character qualifier, if present. chigh = 0 is a default
2193 that signals that the string length = string_length. */
2194
2195 clow = 1;
2196 chigh = 0;
2197
2198 if (c == '(' && nl->type == GFC_DTYPE_CHARACTER)
2199 {
2200 descriptor_dimension chd[1] = { {1, clow, nl->string_length} };
2201 nml_loop_spec ind[1] = { {1, clow, nl->string_length, 1} };
2202
2203 if (nml_parse_qualifier (chd, ind, 1) == FAILURE)
2204 {
2205 st_sprintf (nml_err_msg, "%s for namelist variable %s",
2206 parse_err_msg, nl->var_name);
2207 goto nml_err_ret;
2208 }
2209
2210 clow = ind[0].start;
2211 chigh = ind[0].end;
2212
2213 if (ind[0].step != 1)
2214 {
2215 st_sprintf (nml_err_msg,
2216 "Bad step in substring for namelist object %s",
2217 nl->var_name);
2218 goto nml_err_ret;
2219 }
2220
2221 c = next_char ();
2222 unget_char (c);
2223 }
2224
2225 /* If a derived type touch its components and restore the root
2226 namelist_info if we have parsed a qualified derived type
2227 component. */
2228
2229 if (nl->type == GFC_DTYPE_DERIVED)
2230 nml_touch_nodes (nl);
2231 if (component_flag)
2232 nl = first_nl;
2233
2234 /*make sure no extraneous qualifiers are there.*/
2235
2236 if (c == '(')
2237 {
2238 st_sprintf (nml_err_msg, "Qualifier for a scalar or non-character"
2239 " namelist object %s", nl->var_name);
2240 goto nml_err_ret;
2241 }
2242
2243 /* According to the standard, an equal sign MUST follow an object name. The
2244 following is possibly lax - it allows comments, blank lines and so on to
2245 intervene. eat_spaces (); c = next_char (); would be compliant*/
2246
2247 free_saved ();
2248
2249 eat_separator ();
2250 if (input_complete)
2251 return SUCCESS;
2252
2253 if (at_eol)
2254 finish_separator ();
2255 if (input_complete)
2256 return SUCCESS;
2257
2258 c = next_char ();
2259
2260 if (c != '=')
2261 {
2262 st_sprintf (nml_err_msg, "Equal sign must follow namelist object name %s",
2263 nl->var_name);
2264 goto nml_err_ret;
2265 }
2266
2267 if (nml_read_obj (nl, 0) == FAILURE)
2268 goto nml_err_ret;
2269
2270 return SUCCESS;
2271
2272 nml_err_ret:
2273
2274 return FAILURE;
2275 }
2276
2277 /* Entry point for namelist input. Goes through input until namelist name
2278 is matched. Then cycles through nml_get_obj_data until the input is
2279 completed or there is an error. */
2280
2281 void
2282 namelist_read (void)
2283 {
2284 char c;
2285
2286 namelist_mode = 1;
2287 input_complete = 0;
2288
2289 if (setjmp (g.eof_jump))
2290 {
2291 generate_error (ERROR_END, NULL);
2292 return;
2293 }
2294
2295 /* Look for &namelist_name . Skip all characters, testing for $nmlname.
2296 Exit on success or EOF. If '?' or '=?' encountered in stdin, print
2297 node names or namelist on stdout. */
2298
2299 find_nml_name:
2300 switch (c = next_char ())
2301 {
2302 case '$':
2303 case '&':
2304 break;
2305
2306 case '=':
2307 c = next_char ();
2308 if (c == '?')
2309 nml_query ('=');
2310 else
2311 unget_char (c);
2312 goto find_nml_name;
2313
2314 case '?':
2315 nml_query ('?');
2316
2317 default:
2318 goto find_nml_name;
2319 }
2320
2321 /* Match the name of the namelist. */
2322
2323 nml_match_name (ioparm.namelist_name, ioparm.namelist_name_len);
2324
2325 if (nml_read_error)
2326 goto find_nml_name;
2327
2328 /* Ready to read namelist objects. If there is an error in input
2329 from stdin, output the error message and continue. */
2330
2331 while (!input_complete)
2332 {
2333 if (nml_get_obj_data () == FAILURE)
2334 {
2335 if (current_unit->unit_number != options.stdin_unit)
2336 goto nml_err_ret;
2337
2338 st_printf ("%s\n", nml_err_msg);
2339 flush (find_unit (options.stderr_unit)->s);
2340 }
2341
2342 }
2343
2344 return;
2345
2346 /* All namelist error calls return from here */
2347
2348 nml_err_ret:
2349
2350 generate_error (ERROR_READ_VALUE , nml_err_msg);
2351 return;
2352 }